Microsoft's new “small” AI team ... TL;DR and podcast 28Jan24

Last update: Monday 1/31/24 

Welcome to our 28Jan24 TL;DR summary + podcast about the past week's top AI story on our "Useful AI News" page ➡ Microsoft creates new team to build “small” AI, 

Click link for podcast (opens in new tab) 

➡ Microsoft creates new “small” AI team         

Click the "start" button when the podcast page is loaded
...  If audio fails to start, or gets stuck, try reloading that page.

TL;DR link ➡ HERE

A. TL;DR ... Top story in past week ... 

Although this edition of our TL;DR only identified one "must-read" top story on our Useful AI News page last week, our readers should understand that our news page also identified other articles that they might also find of interest. For example:

• The U.S. Department of Commerce announced its intention to impose a requirement that AI companies notify the U.S. government of their intentions to develop language models that exceeded a specified size before they developed those models. Of course this restriction was conceived before it was generally recognized that small language models had surprising power. So small models, smaller than the required notification level, might also pose unacceptable risks.

• Microsoft announced a pro version of Microsoft Copilot that would provide support for the apps in its productivity suite and would be available to all users who paid a monthly subscription fee
  
  
• Google announced Lumiere, a space-time diffusion model for generating realistic AI videos
  
  
• Stability AI announced a small language model (1.6 billion parameters) that's smaller than Microsoft's Phi-2 (2.7 bilion parameters)

Now let's turn to our top story

1) Microsoft creates a new small language model (SLM) team 

This story was reported exclusively in a tech publication called "The Information", with all but the first few words of the article posted behind an expensive paywall. 

Another publication, Windows Central, published a summary of the paywalled piece. A third tech publication, The Verge, published a less extensive summary that was consistent with the summary posted by Windows Central. Fortunately, the high significance of Microsoft's initiative only depends on two key points:

• Microsoft has formed a team of some of its best techs to develop small language models (SLMs).
  
• The team will report directly to Kevin Scott, Microsoft's chief technology officer (CTO), the same visionary top level manager who managed Microsoft's partnership with OpenAI that resulted in the successful transfer and transformation of OpenAI's large language model technology into copilots for Microsoft's office productivity apps.

That's all we need to know. Why? Because Microsoft's recently announced small language model, Phi-2, has already provided Microsoft with proof-of-concept that SLM's trained on high quality data have surprising power that is comparable to the power of LLM's that are 500 times as large. 

It costs Microsoft billions to fund OpenAI's development of its large language GPT models and is costing billions to operate. If powerful SLMs prove to only cost millions to develop and millions to operate, they will provide Microsoft with more customers, more revenue, and more profits.

That Microsoft should decide to quietly launch a "Manhattan Project" to develop powerful SLMs using its own in-house tech talent, instead of subcontracting this development to OpenAI, is to be expected. The stunning instability shown by OpenAI during the "AI Big Bang" weekend in late November 2023, when OpenAI's CEO Sam Altman was abruptly fired without any warning then rehired a few days later, might have had catastrophic consequences for Microsoft. Fortunately for Microsoft, this catastrophe was averted by the masterful management of the crisis by Satya Nadella, Microsoft's CEO.

B. Top story in past week ... out of 9 new stories 

1. Microsoft
   "Exclusive: Microsoft has created a new team to build “small” AI that’s cheaper than OpenAI’s.", Aaron Holmes, The Information, 1/23/24 ***  ... Link now presents the full paywalled article for 3 free reads 1/31/24
   
   -- An extensive summary of this "exclusive" was published by Windows Central. A brief summary also appeared in The Verge
   
   

C. SLMs and other background for understanding Microsoft's "Manhattan Project"

• "Textbooks Are All You Need II: phi-1.5 technical report", Yuanzhi Li, Sébastien Bubeck, Ronen Eldan, Allie Del Giorno, Suriya Gunasekar, Yin Tat Lee, Microsoft Research, September 2023
• "Phi-2: The surprising power of small language models", Mojan Javaheripi and Sébastien Bubeck , Microsoft Research, 12/12/24
• "The Inside Story of Microsoft’s Partnership with OpenAI", Charles Duhigg, The New Yorker, 12/1/23 
• "Exploring The Future: 5 Cutting-Edge Generative AI Trends In 2024", Janakiram MSV, Forbes, 1/2/24

D. ➡ Dozen Basic AI FAQs

This page contains links to responses by Google's Bard chatbot running Gemini Pro to 12 questions that should be asked more frequently, but aren't. As consequence, too many readily understood AI terms have become meaningless buzzwords in the media.